1 //===- IndVarSimplify.cpp - Induction Variable Elimination ----------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This transformation analyzes and transforms the induction variables (and
11 // computations derived from them) into simpler forms suitable for subsequent
12 // analysis and transformation.
14 // This transformation makes the following changes to each loop with an
15 // identifiable induction variable:
16 // 1. All loops are transformed to have a SINGLE canonical induction variable
17 // which starts at zero and steps by one.
18 // 2. The canonical induction variable is guaranteed to be the first PHI node
19 // in the loop header block.
20 // 3. Any pointer arithmetic recurrences are raised to use array subscripts.
22 // If the trip count of a loop is computable, this pass also makes the following
24 // 1. The exit condition for the loop is canonicalized to compare the
25 // induction value against the exit value. This turns loops like:
26 // 'for (i = 7; i*i < 1000; ++i)' into 'for (i = 0; i != 25; ++i)'
27 // 2. Any use outside of the loop of an expression derived from the indvar
28 // is changed to compute the derived value outside of the loop, eliminating
29 // the dependence on the exit value of the induction variable. If the only
30 // purpose of the loop is to compute the exit value of some derived
31 // expression, this transformation will make the loop dead.
33 // This transformation should be followed by strength reduction after all of the
34 // desired loop transformations have been performed. Additionally, on targets
35 // where it is profitable, the loop could be transformed to count down to zero
36 // (the "do loop" optimization).
38 //===----------------------------------------------------------------------===//
40 #define DEBUG_TYPE "indvars"
41 #include "llvm/Transforms/Scalar.h"
42 #include "llvm/BasicBlock.h"
43 #include "llvm/Constants.h"
44 #include "llvm/Instructions.h"
45 #include "llvm/Type.h"
46 #include "llvm/Analysis/ScalarEvolutionExpander.h"
47 #include "llvm/Analysis/LoopInfo.h"
48 #include "llvm/Analysis/LoopPass.h"
49 #include "llvm/Support/CFG.h"
50 #include "llvm/Support/Compiler.h"
51 #include "llvm/Support/Debug.h"
52 #include "llvm/Support/GetElementPtrTypeIterator.h"
53 #include "llvm/Transforms/Utils/Local.h"
54 #include "llvm/Support/CommandLine.h"
55 #include "llvm/ADT/SmallVector.h"
56 #include "llvm/ADT/SetVector.h"
57 #include "llvm/ADT/SmallPtrSet.h"
58 #include "llvm/ADT/Statistic.h"
61 STATISTIC(NumRemoved , "Number of aux indvars removed");
62 STATISTIC(NumPointer , "Number of pointer indvars promoted");
63 STATISTIC(NumInserted, "Number of canonical indvars added");
64 STATISTIC(NumReplaced, "Number of exit values replaced");
65 STATISTIC(NumLFTR , "Number of loop exit tests replaced");
68 class VISIBILITY_HIDDEN IndVarSimplify : public LoopPass {
74 static char ID; // Pass identification, replacement for typeid
75 IndVarSimplify() : LoopPass(&ID) {}
77 bool runOnLoop(Loop *L, LPPassManager &LPM);
78 bool doInitialization(Loop *L, LPPassManager &LPM);
79 virtual void getAnalysisUsage(AnalysisUsage &AU) const {
80 AU.addRequired<ScalarEvolution>();
81 AU.addRequiredID(LCSSAID);
82 AU.addRequiredID(LoopSimplifyID);
83 AU.addRequired<LoopInfo>();
84 AU.addPreservedID(LoopSimplifyID);
85 AU.addPreservedID(LCSSAID);
91 void EliminatePointerRecurrence(PHINode *PN, BasicBlock *Preheader,
92 SmallPtrSet<Instruction*, 16> &DeadInsts);
93 void LinearFunctionTestReplace(Loop *L, SCEVHandle IterationCount, Value *IndVar,
94 BasicBlock *ExitingBlock,
96 SCEVExpander &Rewriter);
97 void RewriteLoopExitValues(Loop *L, SCEV *IterationCount);
99 void DeleteTriviallyDeadInstructions(SmallPtrSet<Instruction*, 16> &Insts);
101 void HandleFloatingPointIV(Loop *L, PHINode *PH,
102 SmallPtrSet<Instruction*, 16> &DeadInsts);
106 char IndVarSimplify::ID = 0;
107 static RegisterPass<IndVarSimplify>
108 X("indvars", "Canonicalize Induction Variables");
110 Pass *llvm::createIndVarSimplifyPass() {
111 return new IndVarSimplify();
114 /// DeleteTriviallyDeadInstructions - If any of the instructions is the
115 /// specified set are trivially dead, delete them and see if this makes any of
116 /// their operands subsequently dead.
117 void IndVarSimplify::
118 DeleteTriviallyDeadInstructions(SmallPtrSet<Instruction*, 16> &Insts) {
119 while (!Insts.empty()) {
120 Instruction *I = *Insts.begin();
122 if (isInstructionTriviallyDead(I)) {
123 for (unsigned i = 0, e = I->getNumOperands(); i != e; ++i)
124 if (Instruction *U = dyn_cast<Instruction>(I->getOperand(i)))
126 SE->deleteValueFromRecords(I);
127 DOUT << "INDVARS: Deleting: " << *I;
128 I->eraseFromParent();
135 /// EliminatePointerRecurrence - Check to see if this is a trivial GEP pointer
136 /// recurrence. If so, change it into an integer recurrence, permitting
137 /// analysis by the SCEV routines.
138 void IndVarSimplify::EliminatePointerRecurrence(PHINode *PN,
139 BasicBlock *Preheader,
140 SmallPtrSet<Instruction*, 16> &DeadInsts) {
141 assert(PN->getNumIncomingValues() == 2 && "Noncanonicalized loop!");
142 unsigned PreheaderIdx = PN->getBasicBlockIndex(Preheader);
143 unsigned BackedgeIdx = PreheaderIdx^1;
144 if (GetElementPtrInst *GEPI =
145 dyn_cast<GetElementPtrInst>(PN->getIncomingValue(BackedgeIdx)))
146 if (GEPI->getOperand(0) == PN) {
147 assert(GEPI->getNumOperands() == 2 && "GEP types must match!");
148 DOUT << "INDVARS: Eliminating pointer recurrence: " << *GEPI;
150 // Okay, we found a pointer recurrence. Transform this pointer
151 // recurrence into an integer recurrence. Compute the value that gets
152 // added to the pointer at every iteration.
153 Value *AddedVal = GEPI->getOperand(1);
155 // Insert a new integer PHI node into the top of the block.
156 PHINode *NewPhi = PHINode::Create(AddedVal->getType(),
157 PN->getName()+".rec", PN);
158 NewPhi->addIncoming(Constant::getNullValue(NewPhi->getType()), Preheader);
160 // Create the new add instruction.
161 Value *NewAdd = BinaryOperator::CreateAdd(NewPhi, AddedVal,
162 GEPI->getName()+".rec", GEPI);
163 NewPhi->addIncoming(NewAdd, PN->getIncomingBlock(BackedgeIdx));
165 // Update the existing GEP to use the recurrence.
166 GEPI->setOperand(0, PN->getIncomingValue(PreheaderIdx));
168 // Update the GEP to use the new recurrence we just inserted.
169 GEPI->setOperand(1, NewAdd);
171 // If the incoming value is a constant expr GEP, try peeling out the array
172 // 0 index if possible to make things simpler.
173 if (ConstantExpr *CE = dyn_cast<ConstantExpr>(GEPI->getOperand(0)))
174 if (CE->getOpcode() == Instruction::GetElementPtr) {
175 unsigned NumOps = CE->getNumOperands();
176 assert(NumOps > 1 && "CE folding didn't work!");
177 if (CE->getOperand(NumOps-1)->isNullValue()) {
178 // Check to make sure the last index really is an array index.
179 gep_type_iterator GTI = gep_type_begin(CE);
180 for (unsigned i = 1, e = CE->getNumOperands()-1;
183 if (isa<SequentialType>(*GTI)) {
184 // Pull the last index out of the constant expr GEP.
185 SmallVector<Value*, 8> CEIdxs(CE->op_begin()+1, CE->op_end()-1);
186 Constant *NCE = ConstantExpr::getGetElementPtr(CE->getOperand(0),
190 Idx[0] = Constant::getNullValue(Type::Int32Ty);
192 GetElementPtrInst *NGEPI = GetElementPtrInst::Create(
194 GEPI->getName(), GEPI);
195 SE->deleteValueFromRecords(GEPI);
196 GEPI->replaceAllUsesWith(NGEPI);
197 GEPI->eraseFromParent();
204 // Finally, if there are any other users of the PHI node, we must
205 // insert a new GEP instruction that uses the pre-incremented version
206 // of the induction amount.
207 if (!PN->use_empty()) {
208 BasicBlock::iterator InsertPos = PN; ++InsertPos;
209 while (isa<PHINode>(InsertPos)) ++InsertPos;
211 GetElementPtrInst::Create(PN->getIncomingValue(PreheaderIdx),
212 NewPhi, "", InsertPos);
213 PreInc->takeName(PN);
214 PN->replaceAllUsesWith(PreInc);
217 // Delete the old PHI for sure, and the GEP if its otherwise unused.
218 DeadInsts.insert(PN);
225 /// LinearFunctionTestReplace - This method rewrites the exit condition of the
226 /// loop to be a canonical != comparison against the incremented loop induction
227 /// variable. This pass is able to rewrite the exit tests of any loop where the
228 /// SCEV analysis can determine a loop-invariant trip count of the loop, which
229 /// is actually a much broader range than just linear tests.
230 void IndVarSimplify::LinearFunctionTestReplace(Loop *L,
231 SCEVHandle IterationCount,
233 BasicBlock *ExitingBlock,
235 SCEVExpander &Rewriter) {
236 // If the exiting block is not the same as the backedge block, we must compare
237 // against the preincremented value, otherwise we prefer to compare against
238 // the post-incremented value.
240 if (ExitingBlock == L->getLoopLatch()) {
241 // What ScalarEvolution calls the "iteration count" is actually the
242 // number of times the branch is taken. Add one to get the number
243 // of times the branch is executed. If this addition may overflow,
244 // we have to be more pessimistic and cast the induction variable
245 // before doing the add.
246 SCEVHandle Zero = SE->getIntegerSCEV(0, IterationCount->getType());
248 SE->getAddExpr(IterationCount,
249 SE->getIntegerSCEV(1, IterationCount->getType()));
250 if ((isa<SCEVConstant>(N) && !N->isZero()) ||
251 SE->isLoopGuardedByCond(L, ICmpInst::ICMP_NE, N, Zero)) {
252 // No overflow. Cast the sum.
253 IterationCount = SE->getTruncateOrZeroExtend(N, IndVar->getType());
255 // Potential overflow. Cast before doing the add.
256 IterationCount = SE->getTruncateOrZeroExtend(IterationCount,
259 SE->getAddExpr(IterationCount,
260 SE->getIntegerSCEV(1, IndVar->getType()));
263 // The IterationCount expression contains the number of times that the
264 // backedge actually branches to the loop header. This is one less than the
265 // number of times the loop executes, so add one to it.
266 CmpIndVar = L->getCanonicalInductionVariableIncrement();
268 // We have to use the preincremented value...
269 IterationCount = SE->getTruncateOrZeroExtend(IterationCount,
274 // Expand the code for the iteration count into the preheader of the loop.
275 BasicBlock *Preheader = L->getLoopPreheader();
276 Value *ExitCnt = Rewriter.expandCodeFor(IterationCount,
277 Preheader->getTerminator());
279 // Insert a new icmp_ne or icmp_eq instruction before the branch.
280 ICmpInst::Predicate Opcode;
281 if (L->contains(BI->getSuccessor(0)))
282 Opcode = ICmpInst::ICMP_NE;
284 Opcode = ICmpInst::ICMP_EQ;
286 DOUT << "INDVARS: Rewriting loop exit condition to:\n"
287 << " LHS:" << *CmpIndVar // includes a newline
289 << (Opcode == ICmpInst::ICMP_NE ? "!=" : "==") << "\n"
290 << " RHS:\t" << *IterationCount << "\n";
292 Value *Cond = new ICmpInst(Opcode, CmpIndVar, ExitCnt, "exitcond", BI);
293 BI->setCondition(Cond);
298 /// RewriteLoopExitValues - Check to see if this loop has a computable
299 /// loop-invariant execution count. If so, this means that we can compute the
300 /// final value of any expressions that are recurrent in the loop, and
301 /// substitute the exit values from the loop into any instructions outside of
302 /// the loop that use the final values of the current expressions.
303 void IndVarSimplify::RewriteLoopExitValues(Loop *L, SCEV *IterationCount) {
304 BasicBlock *Preheader = L->getLoopPreheader();
306 // Scan all of the instructions in the loop, looking at those that have
307 // extra-loop users and which are recurrences.
308 SCEVExpander Rewriter(*SE, *LI);
310 // We insert the code into the preheader of the loop if the loop contains
311 // multiple exit blocks, or in the exit block if there is exactly one.
312 BasicBlock *BlockToInsertInto;
313 SmallVector<BasicBlock*, 8> ExitBlocks;
314 L->getUniqueExitBlocks(ExitBlocks);
315 if (ExitBlocks.size() == 1)
316 BlockToInsertInto = ExitBlocks[0];
318 BlockToInsertInto = Preheader;
319 BasicBlock::iterator InsertPt = BlockToInsertInto->getFirstNonPHI();
321 bool HasConstantItCount = isa<SCEVConstant>(IterationCount);
323 SmallPtrSet<Instruction*, 16> InstructionsToDelete;
324 std::map<Instruction*, Value*> ExitValues;
326 // Find all values that are computed inside the loop, but used outside of it.
327 // Because of LCSSA, these values will only occur in LCSSA PHI Nodes. Scan
328 // the exit blocks of the loop to find them.
329 for (unsigned i = 0, e = ExitBlocks.size(); i != e; ++i) {
330 BasicBlock *ExitBB = ExitBlocks[i];
332 // If there are no PHI nodes in this exit block, then no values defined
333 // inside the loop are used on this path, skip it.
334 PHINode *PN = dyn_cast<PHINode>(ExitBB->begin());
337 unsigned NumPreds = PN->getNumIncomingValues();
339 // Iterate over all of the PHI nodes.
340 BasicBlock::iterator BBI = ExitBB->begin();
341 while ((PN = dyn_cast<PHINode>(BBI++))) {
343 // Iterate over all of the values in all the PHI nodes.
344 for (unsigned i = 0; i != NumPreds; ++i) {
345 // If the value being merged in is not integer or is not defined
346 // in the loop, skip it.
347 Value *InVal = PN->getIncomingValue(i);
348 if (!isa<Instruction>(InVal) ||
349 // SCEV only supports integer expressions for now.
350 !isa<IntegerType>(InVal->getType()))
353 // If this pred is for a subloop, not L itself, skip it.
354 if (LI->getLoopFor(PN->getIncomingBlock(i)) != L)
355 continue; // The Block is in a subloop, skip it.
357 // Check that InVal is defined in the loop.
358 Instruction *Inst = cast<Instruction>(InVal);
359 if (!L->contains(Inst->getParent()))
362 // We require that this value either have a computable evolution or that
363 // the loop have a constant iteration count. In the case where the loop
364 // has a constant iteration count, we can sometimes force evaluation of
365 // the exit value through brute force.
366 SCEVHandle SH = SE->getSCEV(Inst);
367 if (!SH->hasComputableLoopEvolution(L) && !HasConstantItCount)
368 continue; // Cannot get exit evolution for the loop value.
370 // Okay, this instruction has a user outside of the current loop
371 // and varies predictably *inside* the loop. Evaluate the value it
372 // contains when the loop exits, if possible.
373 SCEVHandle ExitValue = SE->getSCEVAtScope(Inst, L->getParentLoop());
374 if (isa<SCEVCouldNotCompute>(ExitValue) ||
375 !ExitValue->isLoopInvariant(L))
381 // See if we already computed the exit value for the instruction, if so,
383 Value *&ExitVal = ExitValues[Inst];
385 ExitVal = Rewriter.expandCodeFor(ExitValue, InsertPt);
387 DOUT << "INDVARS: RLEV: AfterLoopVal = " << *ExitVal
388 << " LoopVal = " << *Inst << "\n";
390 PN->setIncomingValue(i, ExitVal);
392 // If this instruction is dead now, schedule it to be removed.
393 if (Inst->use_empty())
394 InstructionsToDelete.insert(Inst);
396 // See if this is a single-entry LCSSA PHI node. If so, we can (and
398 // the PHI entirely. This is safe, because the NewVal won't be variant
399 // in the loop, so we don't need an LCSSA phi node anymore.
401 SE->deleteValueFromRecords(PN);
402 PN->replaceAllUsesWith(ExitVal);
403 PN->eraseFromParent();
410 DeleteTriviallyDeadInstructions(InstructionsToDelete);
413 bool IndVarSimplify::doInitialization(Loop *L, LPPassManager &LPM) {
416 // First step. Check to see if there are any trivial GEP pointer recurrences.
417 // If there are, change them into integer recurrences, permitting analysis by
418 // the SCEV routines.
420 BasicBlock *Header = L->getHeader();
421 BasicBlock *Preheader = L->getLoopPreheader();
422 SE = &LPM.getAnalysis<ScalarEvolution>();
424 SmallPtrSet<Instruction*, 16> DeadInsts;
425 for (BasicBlock::iterator I = Header->begin(); isa<PHINode>(I); ++I) {
426 PHINode *PN = cast<PHINode>(I);
427 if (isa<PointerType>(PN->getType()))
428 EliminatePointerRecurrence(PN, Preheader, DeadInsts);
430 HandleFloatingPointIV(L, PN, DeadInsts);
433 if (!DeadInsts.empty())
434 DeleteTriviallyDeadInstructions(DeadInsts);
439 /// getEffectiveIndvarType - Determine the widest type that the
440 /// induction-variable PHINode Phi is cast to.
442 static const Type *getEffectiveIndvarType(const PHINode *Phi) {
443 const Type *Ty = Phi->getType();
445 for (Value::use_const_iterator UI = Phi->use_begin(), UE = Phi->use_end();
447 const Type *CandidateType = NULL;
448 if (const ZExtInst *ZI = dyn_cast<ZExtInst>(UI))
449 CandidateType = ZI->getDestTy();
450 else if (const SExtInst *SI = dyn_cast<SExtInst>(UI))
451 CandidateType = SI->getDestTy();
453 CandidateType->getPrimitiveSizeInBits() >
454 Ty->getPrimitiveSizeInBits())
461 /// isOrigIVAlwaysNonNegative - Analyze the original induction variable
462 /// in the loop to determine whether it would ever have a negative
465 /// TODO: This duplicates a fair amount of ScalarEvolution logic.
466 /// Perhaps this can be merged with ScalarEvolution::getIterationCount.
468 static bool isOrigIVAlwaysNonNegative(const Loop *L,
469 const Instruction *OrigCond) {
470 // Verify that the loop is sane and find the exit condition.
471 const ICmpInst *Cmp = dyn_cast<ICmpInst>(OrigCond);
472 if (!Cmp) return false;
474 // For now, analyze only SLT loops for signed overflow.
475 if (Cmp->getPredicate() != ICmpInst::ICMP_SLT) return false;
477 // Get the increment instruction. Look past SExtInsts if we will
478 // be able to prove that the original induction variable doesn't
479 // undergo signed overflow.
480 const Value *OrigIncrVal = Cmp->getOperand(0);
481 const Value *IncrVal = OrigIncrVal;
482 if (SExtInst *SI = dyn_cast<SExtInst>(Cmp->getOperand(0))) {
483 if (!isa<ConstantInt>(Cmp->getOperand(1)) ||
484 !cast<ConstantInt>(Cmp->getOperand(1))->getValue()
485 .isSignedIntN(IncrVal->getType()->getPrimitiveSizeInBits()))
487 IncrVal = SI->getOperand(0);
490 // For now, only analyze induction variables that have simple increments.
491 const BinaryOperator *IncrOp = dyn_cast<BinaryOperator>(IncrVal);
493 IncrOp->getOpcode() != Instruction::Add ||
494 !isa<ConstantInt>(IncrOp->getOperand(1)) ||
495 !cast<ConstantInt>(IncrOp->getOperand(1))->equalsInt(1))
498 // Make sure the PHI looks like a normal IV.
499 const PHINode *PN = dyn_cast<PHINode>(IncrOp->getOperand(0));
500 if (!PN || PN->getNumIncomingValues() != 2)
502 unsigned IncomingEdge = L->contains(PN->getIncomingBlock(0));
503 unsigned BackEdge = !IncomingEdge;
504 if (!L->contains(PN->getIncomingBlock(BackEdge)) ||
505 PN->getIncomingValue(BackEdge) != IncrOp)
508 // For now, only analyze loops with a constant start value, so that
509 // we can easily determine if the start value is non-negative and
510 // not a maximum value which would wrap on the first iteration.
511 const Value *InitialVal = PN->getIncomingValue(IncomingEdge);
512 if (!isa<ConstantInt>(InitialVal) ||
513 cast<ConstantInt>(InitialVal)->getValue().isNegative() ||
514 cast<ConstantInt>(InitialVal)->getValue().isMaxSignedValue())
517 // The original induction variable will start at some non-negative
518 // non-max value, it counts up by one, and the loop iterates only
519 // while it remans less than (signed) some value in the same type.
520 // As such, it will always be non-negative.
524 bool IndVarSimplify::runOnLoop(Loop *L, LPPassManager &LPM) {
525 LI = &getAnalysis<LoopInfo>();
526 SE = &getAnalysis<ScalarEvolution>();
529 BasicBlock *Header = L->getHeader();
530 BasicBlock *ExitingBlock = L->getExitingBlock();
531 SmallPtrSet<Instruction*, 16> DeadInsts;
533 // Verify the input to the pass in already in LCSSA form.
534 assert(L->isLCSSAForm());
536 // Check to see if this loop has a computable loop-invariant execution count.
537 // If so, this means that we can compute the final value of any expressions
538 // that are recurrent in the loop, and substitute the exit values from the
539 // loop into any instructions outside of the loop that use the final values of
540 // the current expressions.
542 SCEVHandle IterationCount = SE->getIterationCount(L);
543 if (!isa<SCEVCouldNotCompute>(IterationCount))
544 RewriteLoopExitValues(L, IterationCount);
546 // Next, analyze all of the induction variables in the loop, canonicalizing
547 // auxillary induction variables.
548 std::vector<std::pair<PHINode*, SCEVHandle> > IndVars;
550 for (BasicBlock::iterator I = Header->begin(); isa<PHINode>(I); ++I) {
551 PHINode *PN = cast<PHINode>(I);
552 if (PN->getType()->isInteger()) { // FIXME: when we have fast-math, enable!
553 SCEVHandle SCEV = SE->getSCEV(PN);
554 // FIXME: It is an extremely bad idea to indvar substitute anything more
555 // complex than affine induction variables. Doing so will put expensive
556 // polynomial evaluations inside of the loop, and the str reduction pass
557 // currently can only reduce affine polynomials. For now just disable
558 // indvar subst on anything more complex than an affine addrec.
559 if (SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(SCEV))
560 if (AR->getLoop() == L && AR->isAffine())
561 IndVars.push_back(std::make_pair(PN, SCEV));
565 // Compute the type of the largest recurrence expression, and collect
566 // the set of the types of the other recurrence expressions.
567 const Type *LargestType = 0;
568 SmallSetVector<const Type *, 4> SizesToInsert;
569 if (!isa<SCEVCouldNotCompute>(IterationCount)) {
570 LargestType = IterationCount->getType();
571 SizesToInsert.insert(IterationCount->getType());
573 for (unsigned i = 0, e = IndVars.size(); i != e; ++i) {
574 const PHINode *PN = IndVars[i].first;
575 SizesToInsert.insert(PN->getType());
576 const Type *EffTy = getEffectiveIndvarType(PN);
577 SizesToInsert.insert(EffTy);
579 EffTy->getPrimitiveSizeInBits() >
580 LargestType->getPrimitiveSizeInBits())
584 // Create a rewriter object which we'll use to transform the code with.
585 SCEVExpander Rewriter(*SE, *LI);
587 // Now that we know the largest of of the induction variables in this loop,
588 // insert a canonical induction variable of the largest size.
590 if (!SizesToInsert.empty()) {
591 IndVar = Rewriter.getOrInsertCanonicalInductionVariable(L,LargestType);
594 DOUT << "INDVARS: New CanIV: " << *IndVar;
597 // If we have a trip count expression, rewrite the loop's exit condition
598 // using it. We can currently only handle loops with a single exit.
599 bool OrigIVAlwaysNonNegative = false;
600 if (!isa<SCEVCouldNotCompute>(IterationCount) && ExitingBlock)
601 // Can't rewrite non-branch yet.
602 if (BranchInst *BI = dyn_cast<BranchInst>(ExitingBlock->getTerminator())) {
603 if (Instruction *OrigCond = dyn_cast<Instruction>(BI->getCondition())) {
604 // Determine if the OrigIV will ever have a non-zero sign bit.
605 OrigIVAlwaysNonNegative = isOrigIVAlwaysNonNegative(L, OrigCond);
607 // We'll be replacing the original condition, so it'll be dead.
608 DeadInsts.insert(OrigCond);
611 LinearFunctionTestReplace(L, IterationCount, IndVar,
612 ExitingBlock, BI, Rewriter);
615 // Now that we have a canonical induction variable, we can rewrite any
616 // recurrences in terms of the induction variable. Start with the auxillary
617 // induction variables, and recursively rewrite any of their uses.
618 BasicBlock::iterator InsertPt = Header->getFirstNonPHI();
620 // If there were induction variables of other sizes, cast the primary
621 // induction variable to the right size for them, avoiding the need for the
622 // code evaluation methods to insert induction variables of different sizes.
623 for (unsigned i = 0, e = SizesToInsert.size(); i != e; ++i) {
624 const Type *Ty = SizesToInsert[i];
625 if (Ty != LargestType) {
626 Instruction *New = new TruncInst(IndVar, Ty, "indvar", InsertPt);
627 Rewriter.addInsertedValue(New, SE->getSCEV(New));
628 DOUT << "INDVARS: Made trunc IV for type " << *Ty << ": "
633 // Rewrite all induction variables in terms of the canonical induction
635 while (!IndVars.empty()) {
636 PHINode *PN = IndVars.back().first;
637 Value *NewVal = Rewriter.expandCodeFor(IndVars.back().second, InsertPt);
638 DOUT << "INDVARS: Rewrote IV '" << *IndVars.back().second << "' " << *PN
639 << " into = " << *NewVal << "\n";
640 NewVal->takeName(PN);
642 /// If the new canonical induction variable is wider than the original,
643 /// and the original has uses that are casts to wider types, see if the
644 /// truncate and extend can be omitted.
645 if (isa<TruncInst>(NewVal))
646 for (Value::use_iterator UI = PN->use_begin(), UE = PN->use_end();
648 if (isa<ZExtInst>(UI) ||
649 (isa<SExtInst>(UI) && OrigIVAlwaysNonNegative)) {
650 Value *TruncIndVar = IndVar;
651 if (TruncIndVar->getType() != UI->getType())
652 TruncIndVar = new TruncInst(IndVar, UI->getType(), "truncindvar",
654 UI->replaceAllUsesWith(TruncIndVar);
655 if (Instruction *DeadUse = dyn_cast<Instruction>(*UI))
656 DeadInsts.insert(DeadUse);
659 // Replace the old PHI Node with the inserted computation.
660 PN->replaceAllUsesWith(NewVal);
661 DeadInsts.insert(PN);
668 // Now replace all derived expressions in the loop body with simpler
670 for (LoopInfo::block_iterator I = L->block_begin(), E = L->block_end();
673 if (LI->getLoopFor(BB) == L) { // Not in a subloop...
674 for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I)
675 if (I->getType()->isInteger() && // Is an integer instruction
677 !Rewriter.isInsertedInstruction(I)) {
678 SCEVHandle SH = SE->getSCEV(I);
679 Value *V = Rewriter.expandCodeFor(SH, I, I->getType());
681 if (isa<Instruction>(V))
683 I->replaceAllUsesWith(V);
693 DeleteTriviallyDeadInstructions(DeadInsts);
694 assert(L->isLCSSAForm());
698 /// Return true if it is OK to use SIToFPInst for an inducation variable
699 /// with given inital and exit values.
700 static bool useSIToFPInst(ConstantFP &InitV, ConstantFP &ExitV,
701 uint64_t intIV, uint64_t intEV) {
703 if (InitV.getValueAPF().isNegative() || ExitV.getValueAPF().isNegative())
706 // If the iteration range can be handled by SIToFPInst then use it.
707 APInt Max = APInt::getSignedMaxValue(32);
708 if (Max.getZExtValue() > static_cast<uint64_t>(abs(intEV - intIV)))
714 /// convertToInt - Convert APF to an integer, if possible.
715 static bool convertToInt(const APFloat &APF, uint64_t *intVal) {
717 bool isExact = false;
718 if (&APF.getSemantics() == &APFloat::PPCDoubleDouble)
720 if (APF.convertToInteger(intVal, 32, APF.isNegative(),
721 APFloat::rmTowardZero, &isExact)
730 /// HandleFloatingPointIV - If the loop has floating induction variable
731 /// then insert corresponding integer induction variable if possible.
733 /// for(double i = 0; i < 10000; ++i)
735 /// is converted into
736 /// for(int i = 0; i < 10000; ++i)
739 void IndVarSimplify::HandleFloatingPointIV(Loop *L, PHINode *PH,
740 SmallPtrSet<Instruction*, 16> &DeadInsts) {
742 unsigned IncomingEdge = L->contains(PH->getIncomingBlock(0));
743 unsigned BackEdge = IncomingEdge^1;
745 // Check incoming value.
746 ConstantFP *InitValue = dyn_cast<ConstantFP>(PH->getIncomingValue(IncomingEdge));
747 if (!InitValue) return;
748 uint64_t newInitValue = Type::Int32Ty->getPrimitiveSizeInBits();
749 if (!convertToInt(InitValue->getValueAPF(), &newInitValue))
752 // Check IV increment. Reject this PH if increement operation is not
753 // an add or increment value can not be represented by an integer.
754 BinaryOperator *Incr =
755 dyn_cast<BinaryOperator>(PH->getIncomingValue(BackEdge));
757 if (Incr->getOpcode() != Instruction::Add) return;
758 ConstantFP *IncrValue = NULL;
759 unsigned IncrVIndex = 1;
760 if (Incr->getOperand(1) == PH)
762 IncrValue = dyn_cast<ConstantFP>(Incr->getOperand(IncrVIndex));
763 if (!IncrValue) return;
764 uint64_t newIncrValue = Type::Int32Ty->getPrimitiveSizeInBits();
765 if (!convertToInt(IncrValue->getValueAPF(), &newIncrValue))
768 // Check Incr uses. One user is PH and the other users is exit condition used
769 // by the conditional terminator.
770 Value::use_iterator IncrUse = Incr->use_begin();
771 Instruction *U1 = cast<Instruction>(IncrUse++);
772 if (IncrUse == Incr->use_end()) return;
773 Instruction *U2 = cast<Instruction>(IncrUse++);
774 if (IncrUse != Incr->use_end()) return;
776 // Find exit condition.
777 FCmpInst *EC = dyn_cast<FCmpInst>(U1);
779 EC = dyn_cast<FCmpInst>(U2);
782 if (BranchInst *BI = dyn_cast<BranchInst>(EC->getParent()->getTerminator())) {
783 if (!BI->isConditional()) return;
784 if (BI->getCondition() != EC) return;
787 // Find exit value. If exit value can not be represented as an interger then
788 // do not handle this floating point PH.
789 ConstantFP *EV = NULL;
790 unsigned EVIndex = 1;
791 if (EC->getOperand(1) == Incr)
793 EV = dyn_cast<ConstantFP>(EC->getOperand(EVIndex));
795 uint64_t intEV = Type::Int32Ty->getPrimitiveSizeInBits();
796 if (!convertToInt(EV->getValueAPF(), &intEV))
799 // Find new predicate for integer comparison.
800 CmpInst::Predicate NewPred = CmpInst::BAD_ICMP_PREDICATE;
801 switch (EC->getPredicate()) {
802 case CmpInst::FCMP_OEQ:
803 case CmpInst::FCMP_UEQ:
804 NewPred = CmpInst::ICMP_EQ;
806 case CmpInst::FCMP_OGT:
807 case CmpInst::FCMP_UGT:
808 NewPred = CmpInst::ICMP_UGT;
810 case CmpInst::FCMP_OGE:
811 case CmpInst::FCMP_UGE:
812 NewPred = CmpInst::ICMP_UGE;
814 case CmpInst::FCMP_OLT:
815 case CmpInst::FCMP_ULT:
816 NewPred = CmpInst::ICMP_ULT;
818 case CmpInst::FCMP_OLE:
819 case CmpInst::FCMP_ULE:
820 NewPred = CmpInst::ICMP_ULE;
825 if (NewPred == CmpInst::BAD_ICMP_PREDICATE) return;
827 // Insert new integer induction variable.
828 PHINode *NewPHI = PHINode::Create(Type::Int32Ty,
829 PH->getName()+".int", PH);
830 NewPHI->addIncoming(ConstantInt::get(Type::Int32Ty, newInitValue),
831 PH->getIncomingBlock(IncomingEdge));
833 Value *NewAdd = BinaryOperator::CreateAdd(NewPHI,
834 ConstantInt::get(Type::Int32Ty,
836 Incr->getName()+".int", Incr);
837 NewPHI->addIncoming(NewAdd, PH->getIncomingBlock(BackEdge));
839 ConstantInt *NewEV = ConstantInt::get(Type::Int32Ty, intEV);
840 Value *LHS = (EVIndex == 1 ? NewPHI->getIncomingValue(BackEdge) : NewEV);
841 Value *RHS = (EVIndex == 1 ? NewEV : NewPHI->getIncomingValue(BackEdge));
842 ICmpInst *NewEC = new ICmpInst(NewPred, LHS, RHS, EC->getNameStart(),
843 EC->getParent()->getTerminator());
845 // Delete old, floating point, exit comparision instruction.
846 EC->replaceAllUsesWith(NewEC);
847 DeadInsts.insert(EC);
849 // Delete old, floating point, increment instruction.
850 Incr->replaceAllUsesWith(UndefValue::get(Incr->getType()));
851 DeadInsts.insert(Incr);
853 // Replace floating induction variable. Give SIToFPInst preference over
854 // UIToFPInst because it is faster on platforms that are widely used.
855 if (useSIToFPInst(*InitValue, *EV, newInitValue, intEV)) {
856 SIToFPInst *Conv = new SIToFPInst(NewPHI, PH->getType(), "indvar.conv",
857 PH->getParent()->getFirstNonPHI());
858 PH->replaceAllUsesWith(Conv);
860 UIToFPInst *Conv = new UIToFPInst(NewPHI, PH->getType(), "indvar.conv",
861 PH->getParent()->getFirstNonPHI());
862 PH->replaceAllUsesWith(Conv);
864 DeadInsts.insert(PH);